Russian Digital Media and Information Ecosystem in Turkey - H. Akın Ünver | EDAM, Oxford CTGA & Kadir Has University
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Cyber Governance and Digital Democracy 2019/1 March 2019 Russian Digital Media and Information Ecosystem in Turkey H. Akın Ünver | EDAM, Oxford CTGA & Kadir Has University
Cyber Governance and Digital Democracy 2019/1 Russian Digital Media and Information Ecosystem in Turkey H. Akın Ünver | EDAM, Oxford CTGA & Kadir Has Üniversitesi INTRODUCTION In recent years, Russian digital information operations, not confined to the country’s official communication policy. including disinformation, fake news, and election meddling These strategies are part of the Russian military doctrine, have assumed prominence in international news and most relevant of which has been the 2010 Military Doctrine scholarly research outlets. A simple Google Trends query of the Russian Federation, which sought to “escalate to de- shows us that ‘fake news’ as a term enters into global escalate”2 tensions encompassing the country’s western mainstream lexicon starting with October 2016, peaking borders. To achieve this, the document advised ‘hybrid in the immediate aftermath of the US Presidential Election war’, which is an umbrella term to define untraceable and in November. Since then, disinformation has been largely largely non-violent tools and methods that complement synonymous with Russian digital information operations in conventional military efforts. The 2010 doctrine was further the West, and a number of empirical research projects have bolstered by the 2013 Gerasimov Doctrine, which, among begun focusing on the impact of information warfare on other things, diagnosed the “blurring the lines between elections and political behavior. the states of war and peace”, adding that “wars are no longer declared and having begun, proceed according Russian media ecosystem in Western democracies, to an unfamiliar template”3. Hybrid war is not a Russian including information and dis-information dynamics, invention, nor is Russia the first state to use non-military are quite well-documented1. This focus owes largely to measures to complement military efforts. Rather, the 2010 increased awareness of election meddling, fake news and doctrine was an acknowledgment of the term ‘hybrid war’, digital spoilers such as trolls and bots that often have real- officially coined first by the USCENTCOM in its analysis of life effects. In addition to other digital contestation types, the 2006 Israel-Hezbollah War4. The use of conventional including cyber warfare, Russian information operations are and unconventional tactics, coupled with the new advances This research has been made possible by funding obtained from the US-based Chrest Foundation for the project “Turkey and Russia: Context and Prospects”. 1 Hunt Allcott and Matthew Gentzkow, “Social Media and Fake News in the 2016 Election,” Working Paper (National Bureau of Economic Research, January 2017), https:// doi.org/10.3386/w23089; Meital Balmas, “When Fake News Becomes Real: Combined Exposure to Multiple News Sources and Political Attitudes of Inefficacy, Alienation, and Cynicism,” Communication Research 41, no. 3 (April 1, 2014): 430–54, https://doi.org/10.1177/0093650212453600; Michael C. Dorf and Sidney Tarrow, “Stings and Scams: ‘Fake News,’ the First Amendment, and the New Activist Journalism,” SSRN Scholarly Paper (Rochester, NY: Social Science Research Network, January 26, 2017), https://papers.ssrn.com/abstract=2906444; David M. J. Lazer et al., “The Science of Fake News,” Science 359, no. 6380 (March 9, 2018): 1094–96, https://doi. org/10.1126/science.aao2998. 2 “Text of Newly-Approved Russian Military Doctrine,” Carnegie Endowment for International Peace, February 5, 2010, https://carnegieendowment.org/2010/02/05/text-of-newly-approved-russian-military-doctrine-pub-40266. 3 Henry Foy, “Valery Gerasimov, the General with a Doctrine for Russia,” Financial Times, September 15, 2017, https://www.ft.com/content/7e14a438-989b-11e7-a652-cde3f882dd7b. 4 Matt M. Matthews, “We Were Caught Unprepared: The 2006 Hezbollah-Israeli War” (Fort Leavenworth, Kansas: US Army Combined Arms Center, 2007), https://www.armyupress.army.mil/Portals/7/combat-studies-institute/csi-books/we-were-caught-unprepared.pdf. 1
Cyber Governance and Digital Democracy 2019/1 in communication technology, widened the battlefield and Facebook, as the Arab Spring and Occupy movements used forced the sides to fight along a broad spectrum of ideas, primarily Twitter to organize and disseminate messages8. images and appearances, all floating in digital space. In The publicly visible 140-character platform architecture the words of Timothy Thomas, Russian high command was of Twitter, combined with its fast media upload system, deeply influenced by the 2006 USCENTCOM report and that rendered it the primary venue for critical information flows analysis had led Moscow’s transition into a new thinking in during emergencies, protests and civil wars. The Syrian terms of how to merge new communication technologies Civil War, conflict in Ukraine and war against ISIS have all with strategic thinking: “a real cognitive war underway in the substantially contributed to the rise of Twitter as the primary ether and media for the hearts and minds of its citizens at emergency-related social media platform9. home and abroad”5. Both the NATO Bi-Strategic Capstone Concept10 and the In many ways, the Internet has become a force domain, 2010 Russian Military Doctrine11 document have underlined just like land, sea and air. In January 2019, the world has the threat of an adversary “with the ability to simultaneously attained 51% Internet penetration, meaning more than half employ conventional and non-conventional means of the world is now online and digitally interconnected6. adaptively in pursuit of their objectives”. These means Foreseeing an inevitable mass global interconnectivity, most were fairly identical in both NATO and Russian military major countries have already set up long-term strategies in documents: nuclear proliferation, terrorism, cybercrime place to situate themselves into a more favorable strategic and cyberwar, organized crime and its role in drugs, arms position in the digital domain. For the rest of the state actors, and human trafficking, migration, ethnic and religious there have been two real wake-up calls to adapt to the conflicts, population conflicts due to resource scarcity and digital medium. The first was the Arab Spring movement that globalization. Both documents also emphasized the digital rocked the MENA capitals through 2010-12 and the second medium as an emerging frontier of political contestation. was the Occupy-inspired or related movements that did NATO followed-up with a 2011 ‘Countering Hybrid Threats’ the same in the West . Both movements demonstrated the 7 experiment to develop a unified alliance strategy against disruptive capacity of social media platforms to circumvent disinformation and media manipulation efforts. This and bypass state surveillance and repression. It is during this was ultimately abandoned due to uneven interest and period that social media has begun to transform. Instagram commitment by the constituent countries12. Compared was launched in October 2010, following Facebook’s to NATO, however, Moscow was quicker to embrace the politically important geotag function via ‘Places’ app in uncertainty of the new information revolution, the hybrid August 2010. Facebook bought Instagram in April 2012 and nature of social media and how its intricate twists and turns WhatsApp in February 2014, turning itself into the biggest could be deployed to support what would later be defined heavyweight in social media. In tandem, Twitter emerged as the ‘sub-threshold warfare strategy’. as a more important political communication alternative to 5 Timothy Thomas, “Russia’s 21st Century Information War: Working to Undermine and Destabilize Populations” (Riga: NATO STRATCOM, 2015), https://www.stratcomcoe.org/timothy-thomas-russias-21st-century-information-war-working-undermine-and-destabilize-populations. 6 Abdi Latif Dahir, “Half the World’s Population Used the Internet in 2018 - ITU — Quartz Africa,” Quartz, December 11, 2018, https://qz.com/africa/1490997/more-than-half-of-worlds-population-using-the-internet-in-2018/. 7 Philip N. Howard et al., “Opening Closed Regimes: What Was the Role of Social Media During the Arab Spring?,” SSRN Scholarly Paper (Rochester, NY: Social Science Research Network, 2011), https://papers.ssrn.com/abstract=2595096. 8 Alexandra Segerberg and W. Lance Bennett, “Social Media and the Organization of Collective Action: Using Twitter to Explore the Ecologies of Two Climate Change Protests,” The Communication Review 14, no. 3 (July 1, 2011): 197–215, https://doi.org/10.1080/10714421.2011.597250; W. Lance Bennett and Alexandra Segerberg, “Digital Media and the Personalization of Collective Action,” Information, Communication & Society 14, no. 6 (September 1, 2011): 770–99, https://doi.org/10.1080/1369118X.2011.579141. 9 Markus Rohde et al., “Out of Syria: Mobile Media in Use at the Time of Civil War,” International Journal of Human–Computer Interaction 32, no. 7 (July 2, 2016): 515–31, https://doi.org/10.1080/10447318.2016.1177300. 10 For full text, see: http://www.act.nato.int/images/stories/events/2010/20100826_bi-sc_cht.pdf 11 For full text English translation, see: https://carnegieendowment.org/files/2010russia_military_doctrine.pdf 12 Michael Aaronson et al., “NATO Countering the Hybrid Threat,” PRISM 2, no. 4 (2011): 111–24. 2
Cyber Governance and Digital Democracy 2019/1 Sub-threshold warfare strategy is a sub-strand within the proclaimed during the early 1950s to deter the Soviet Union wider umbrella term of ‘hybrid warfare’, which seeks to through nuclear means, without spending excessively on conduct confrontational and combative operations without conventional forces. The second offset strategy was during triggering the NATO Article 5 obligations or a direct military the 1975-89 period when the United States attempted to retaliation by a NATO country13. This strategy builds upon the pursue technological deterrence against the Warsaw Pact late-Soviet strategy of ‘active measures’, which deployed a to mask NATO’s comparative conventional disadvantage combination of informatics and political framing mechanisms in Eastern Europe. Finally, the third US offset strategy, to divert, distract and mislead institutions and agencies in which Russia is currently challenging in direct terms, was Western countries . As outlined by former KGB Director of 14 announced in 2014 to bolster US capabilities against anti- Foreign Counterespionage Oleg Kalugin, ‘active measures’ access, area-denial (A2-AD) systems developed by Russia worked by creating several layers of separation between and China17. This implied bolstering US cyber surveillance, the perpetrating agency or figures, rendering the operation intelligence, digital media and stealth platforms to preserve virtually untraceable back to Moscow15. Following decades its informatics upper hand in Eastern Europe, especially of iterations, ‘active measures strategy’ has evolved into its along the Russian border. From Russia’s point of view, such modern form - sub-threshold warfare – which defines the US-origin measures targeted ethnic and religious fault lines in sum of non-violent and obstructionist tactics of Russia’s former Soviet countries, to uproot pro-Russian governments hybrid warfare operations within NATO countries. and leaders from power18. In the same vein, Russia’s sub- threshold strategy is a mirror image of US offset strategies. Russia did not invent the sub-threshold warfare, however. By using digital media and informatics tools, Russia seeks It is the Russian response to the American ‘offset strategy’, to offset NATO’s technological and military strength, driving which seeks to alter the balance of power in an unfavorable wedges within and around NATO countries without triggering standoff through creating a new standoff in a more favorable their conventional defense mechanisms19. contestation area . The first American offset strategy was 16 13 Alexander Lanoszka, “Russian Hybrid Warfare and Extended Deterrence in Eastern Europe,” International Affairs 92, no. 1 (January 1, 2016): 175–95, https://doi.org/10.1111/1468-2346.12509. 14 Steve Abrams, “Beyond Propaganda: Soviet Active Measures in Putin’s Russia,” Connections 15, no. 1 (2016): 5–31. 15 Evan Osnos, David Remnick, and Joshua Yaffa, “Trump, Putin, and the New Cold War,” February 24, 2017, https://www.newyorker.com/magazine/2017/03/06/trump-putin-and-the-new-cold-war. 16 Daniel Fiott, “Europe and the Pentagon’s Third Offset Strategy,” The RUSI Journal 161, no. 1 (January 2, 2016): 26–31, https://doi.org/10.1080/03071847.2016.1152118. 17 Luis Simón, “The ‘Third’ US Offset Strategy and Europe’s ‘Anti-Access’ Challenge,” Journal of Strategic Studies 39, no. 3 (April 15, 2016): 417–45, https://doi.org/10.1080/01402390.2016.1163260. 18 Daniel Fiott, “A Revolution Too Far? US Defence Innovation, Europe and NATO’s Military-Technological Gap,” Journal of Strategic Studies 40, no. 3 (April 16, 2017): 417–37, https://doi.org/10.1080/01402390.2016.1176565. 19 Bettina Renz, “Russia and ‘hybrid Warfare,’” Contemporary Politics 22, no. 3 (July 2, 2016): 283–300, https://doi.org/10.1080/13569775.2016.1201316; Sascha Dov Bachmann and Hakan Gunneriusson, “Russia’s Hybrid Warfare in the East: The Integral Nature of the Information Sphere,” SSRN Scholarly Paper (Rochester, NY: Social Science Research Network, October 7, 2015), https://papers.ssrn.com/abstract=2670527. 3
Cyber Governance and Digital Democracy 2019/1 Major Cases of Disinformation and Countermeasures in the West The highest-profile accusations of Russian meddling in the racial and religious tensions and fears among the American West came after the 2016 US General Election. The US right-wing voters, and fed both inaccurate and fabricated Department of Homeland Security (DHS) and the Office of content to channel those tensions and grievances into pro- the Director of National Intelligence (ODNI) have stated that Trump electoral behavior. According to the report, not only Russian President Vladimir Putin had personally ordered a did this targeted disinformation effort contribute substantially large-scale and high-level ‘influence campaign’ to increase to Donald Trump’s victory, but also continued to bolster his the chances of Donald Trump’s victory in the election . This 20 digital popularity during contested decision phases of his was done, according to US security agencies, through a Presidency. Russian military intelligence (GRU) led effort in hacking the Democratic National Committee servers, as well as John What is interesting from the researcher’s point of view, is that Podesta’s – the director of the Hillary Clinton campaign21 - Russian disinformation ecosystem in the US is extremely account. Later in January 2017, Director of the US National easy to spot and map-out, as the OII report demonstrates Intelligence James Clapper asserted in a testimony that how 99% of all engagement related to the pro-Russian Russia was also involved in a coordinated and state-led and Russian content (likes, shares, retweets, comments) ‘fake news campaign’, disseminated across web-based originated from only 20 accounts on Twitter and Facebook, news and social media platforms. All of this led to the all controlled by the IRA, containing account names such as well-known ‘Mueller investigation’, conducted by the US “Being Patriotic,” “Heart of Texas,” “Blacktivist” and “Army Department of Justice Special Counsel Robert Mueller since of Jesus”. An overwhelming majority of these accounts May 2017, exploring the extent to which Russia and pro- share links from explicitly Russian news websites such Russian networks have been involved in the 2016 election . 22 as Russia Today, Sputnik and RIA Novosti, making web A major sub-thread of the investigation concerns the extent domain tracking one of the most common analytical tools of Russian digital media operations in the United States that to identify a pro-Russian network in a large data cluster. go beyond more direct attacks such as hacking. In other words, Russia hasn’t spent much effort in trying to conceal its digital influence operations in the United States, In mid-December 2018, two major empirical studies were and most of these accounts still exist in the American submitted to the US Senate Intelligence Committee that information ecosystem with different names, continuing to explored the measurable impact of Russian disinformation shape opinion within the far-right information networks. In operations in the US elections23. One of these reports, analytical terms, all of these factors render the US one of the conducted by the Oxford Internet Institute’s (OII) easiest cases to study Russian disinformation, as influence Computational Propaganda Project, outlines the extent networks and content are still very much ‘out there’ and to which the Russian Internet Research Agency (IRA) has can be extracted through a very small sample. Even more used targeted disinformation, bots and trolls to divide the US interesting from the researcher’s point of view, Russian public opinion into politically polarized interest groups for disinformation operations have become even more brazen, targeted manipulation . Specifically, the report outlines how 24 direct and identifiable after they were spotted by the Mueller Russian efforts pinpointed and exacerbated existing social, investigation, rendering their identification and network far 20 Karen Yourish and Troy Griggs, “8 U.S. Intelligence Groups Blame Russia for Meddling, but Trump Keeps Clouding the Picture,” The New York Times, July 16, 2018, sec. U.S., https://www.nytimes.com/interactive/2018/07/16/us/elections/russian-interference-statements-comments.html, https://www.nytimes.com/interactive/2018/07/16/us/elections/russian-interference-statements-comments.html. 21 David E. Sanger and Charlie Savage, “U.S. Says Russia Directed Hacks to Influence Elections,” The New York Times, December 21, 2017, sec. U.S., https://www.nytimes.com/2016/10/08/us/politics/us-formally-accuses-russia-of-stealing-dnc-emails.html. 22 Jason Breslow, “All The Criminal Charges To Emerge So Far From Robert Mueller’s Investigation,” NPR.org, December 9, 2018, https://www.npr.org/2018/12/09/643444815/all-the-criminal-charges-to-emerge-so-far-from-robert-muellers-investigation. 23 Scott Shane, “Five Takeaways From New Reports on Russia’s Social Media Operations,” The New York Times, December 18, 2018, sec. U.S., https://www.nytimes.com/2018/12/17/us/politics/takeaways-russia-social-media-operations.html. 24 Philip N. Howard et al., “The IRA and Political Polarization in the United States, 2012-2018,” Computational Propaganda Research Project (Oxford, UK: Oxford Internet Institute, December 2018), https://comprop.oii.ox.ac.uk/research/ira-political-polarization/. 4
Cyber Governance and Digital Democracy 2019/1 easier to map out25. As this research report will show, this a) retaliate against NATO-led ethnic and religious division is quite 180-degrees the opposite of what we observed in strategies in and around Russia, by dividing Western Turkey, where pro-Russian information network is elusive, nations electorally and socially through information warfare, withdrawn and largely dormant, leading to excruciating without triggering Article #5, and b) to ride the wave of rising research difficulties in mapping out the true extent of similar far-right and left-wing populism to maximize the effect of information operations. polarization operations29. Compared to the first wave, the second wave has been far more successful in terms of its Given the influence and success of Russian information goal of incapacitating NATO countries’ collective defense operations in the United States, most analysts have mechanisms and strategic coherence. Beginning with turned to other NATO countries to see whether Russian- 2014, Russia has specifically targeted political parties and affiliated individuals or networks are involved in elections or movements that contributed to the polarization in European contested political episodes. Russian information footprint countries and also those that posed a more existential is more visible in some European countries than others criticism of the political system, rather than individual and came in two major waves . The first wave, 1991- 26 political parties30. 2004, focused primarily on former Soviet states and Cold War frontier countries to promote pro-Moscow political In France, Kremlin has partnered with Front National and candidates and shift the public debate into a form more has verifiably conducted a range of digital warfare attempts, palatable to Kremlin27. As demonstrated in Way (2015), including the hacking and leakage of Emmanuel Macron’s where the anti-Russian candidate is a democrat, Russia campaign team data31. In the UK, pro-Russian accounts promoted more autocratic messaging and information were heavily involved in the Brexit referendum through cyber- in digital media channels, whereas if the anti-Russian attacks, digital disinformation campaigns and targeted candidate was authoritarian, Russian messaging promoted political advertisements to steer the direction of the vote in pluralism, change, and democracy28. The first wave also favor of the ‘Leave’ campaign32. In Germany, Russia has had a generally low level of success. The second wave of been involved with phishing attacks against political parties Russian information operations began in 2014, right after the and campaigns that are pro-EU, including a 2015 hacking United States declared its Third Offset Strategy in November of the German Bundestag, stealing 16 gigabytes of emails 2014, and continues until today. The second wave directly (although these emails weren’t leaked)33. Similar digital targeted core NATO countries with two priorities in mind; disinformation, phishing and campaign hacking cases are 25 Jane Mayer, “How Russia Helped Swing the Election for Trump,” The New Yorker, September 24, 2018, https://www.newyorker.com/magazine/2018/10/01/how-russia-helped-to-swing-the-election-for-trump. 26 Naja Bentzen, “Foreign Influence Operations in the EU - Think Tank” (Brussels: European Parliament, 2018), http://www.europarl.europa.eu/thinktank/en/document.html?reference=EPRS_BRI%282018%29625123. 27 Charlotte Wagnsson and Maria Hellman, “Normative Power Europe Caving In? EU under Pressure of Russian Information Warfare,” JCMS: Journal of Common Market Studies 56, no. 5 (2018): 1161–77, https://doi.org/10.1111/jcms.12726. 28 Lucan A. Way, “The Limits of Autocracy Promotion: The Case of Russia in the ‘near Abroad,’” European Journal of Political Research 54, no. 4 (2015): 691–706, https://doi.org/10.1111/1475-6765.12092. 29 Peter Pomerantsev, “Authoritarianism Goes Global (II): The Kremlin’s Information War,” Journal of Democracy 26, no. 4 (October 19, 2015): 40–50, https://doi.org/10.1353/jod.2015.0074. 30 Alina Polyakova, “Strange Bedfellows: Putin and Europe’s Far Right,” World Affairs 177, no. 3 (2014): 36–40. 31 Alex Hern, “Macron Hackers Linked to Russian-Affiliated Group behind US Attack,” The Guardian, May 8, 2017, sec. World news, https://www.theguardian.com/world/2017/may/08/macron-hackers-linked-to-russian-affiliated-group-behind-us-attack. 32 Patrick Wintour, “Russian Bid to Influence Brexit Vote Detailed in New US Senate Report,” The Guardian, January 10, 2018, sec. World news, https://www.theguardian.com/world/2018/jan/10/russian-influence-brexit-vote-detailed-us-senate-report. 33 Paul Carrel and Andrea Shalal, “Germany Says Its Government Computers Secure after ‘Isolated’ Hack,” Reuters, February 28, 2018, https://www.reuters.com/article/us-germany-cyber-russia-idUSKCN1GC2HZ. 5
Cyber Governance and Digital Democracy 2019/1 also documented in Italy34, Netherlands35 and Sweden36. to mitigate the value of ‘real’ documents that Russia may By December 2018, the European Union has declared have extracted through its hacks. Macron team also hired a belated ‘War Against Disinformation’ in preparation for three IT lawyers, each tasked with handling different aspects renewed Russian meddling efforts in a series of upcoming of disinformation campaigns during the election. Le Monde elections across Europe in 2019 (including the European has published a list of news websites before the election day, Parliament election) . Some of the proposed Europe-wide 37 ranking them according to their reliability40. Also, 30 media efforts are the establishment of a rapid warning system to outlets in France partnered with Google to build a networked recognize, isolate and remove fake or manipulated digital fact-checking initiative called CrossCheck. All of this meant content, establishment of a new ‘digital contract’ with the that Russian meddling in French efforts failed, because main social media platforms Twitter, Facebook, YouTube France launched a truly national, trans-partisan, inclusive and Instagram to ‘get serious’ about tackling disinformation and heavily institutionalized framework to minimize the during key events and set up a European fact-checking damage caused by Russian digital information operations. It network of local verifiers to spot disinformation attempts also used a hybrid strategy of both autonomous, citizen-led in real-time. At the national level, countries have begun fact-checking efforts and strategically deployed deliberate formulating largely converging strategies to combat external disinformation against Russian hackers. information operations. Because the United Kingdom had already suffered France mobilized its intelligence agencies in the run-up to from Russian disinformation attempts during the Brexit its 2017 elections. The National Cybersecurity Agency of Referendum, its current strategy is structured to learn France (ANSSI) has produced a cybersecurity handbook, from those mistakes retrospectively. Like France, Britain’s including a beginner’s introduction to DDoS (Distributed National Cyber Security Center (NCSC) is bringing British Denial of Service) attacks, with follow-up briefings for all political parties together to brief them about the potential political parties38. According to Carnegie Endowment, Marine vulnerabilities of digital information systems and how to Le Pen’s Front National was the only party to be absent secure their networks against phishing attacks41. NCSC from all of these briefings39. This strategy worked, because also published technical primers for politicians on more 2017 French elections became one of the best-documented advanced topics in cybersecurity, disinformation and cases of failed Russian election meddling. Expecting a digital leaks42. Because voting in Britain is complex due hack, Emmanuel Macron’s campaign team hired an IT team to the absence of electronic voting and the responsibility specializing in digital disinformation and generated a digital of organizing voting at the district level, Britain is relatively file storage system that is designed to feed Russia its own more immune to direct digital election meddling. Rather, medicine: deliberately fabricated false campaign documents Britain is more exposed to digital disinformation and opinion 34 Stephanie Kirchgaessner, “Russia Suspected over Hacking Attack on Italian Foreign Ministry,” The Guardian, February 10, 2017, sec. World news, https://www.theguardian.com/world/2017/feb/10/russia-suspected-over-hacking-attack-on-italian-foreign-ministry. 35 Patrick Wintour and Andrew Roth, “Russia Summons Dutch Ambassador over Hacking Revelations,” The Guardian, October 8, 2018, sec. World news, https://www.theguardian.com/world/2018/oct/08/russia-summons-dutch-ambassador-over-hacking-revelations. 36 Erik Brattberg and Tim Maurer, “How Sweden Is Preparing for Russia to Hack Its Election,” May 31, 2018, sec. World, https://www.bbc.com/news/world-44070469. 37 Daniel Boffey, “EU Raises Funds to Fight ‘Disinformation War’ with Russia,” The Guardian, December 5, 2018, sec. World news, https://www.theguardian.com/world/2018/dec/05/eu-disinformation-war-russia-fake-news. 38 Heather A. Conley and Jean-Baptiste Jeangène Wilmer, “Successfully Countering Russian Electoral Interference,” CSIS Briefs (Washington DC: Center for Strategic and International Studies, June 21, 2018), https://www.csis.org/analysis/successfully-countering-russian-electoral-interference. 39 Erik Brattberg and Tim Maurer, “Russian Election Interference: Europe’s Counter to Fake News and Cyber Attacks” (Washington DC: Carnegie Endowment for International Peace, May 23, 2018), https://carnegieendowment.org/2018/05/23/russian-election-interference-europe-s-counter-to-fake-news-and-cyber-attacks-pub-76435. 40 Laura Daniels, “How Russia Hacked the French Election,” POLITICO, April 23, 2017, https://www.politico.eu/article/france-election-2017-russia-hacked-cyberattacks/. 41 Oscar Williams, “Russia Is Targeting UK Infrastructure through Supply Chains, NCSC Warns,” New Statesman, April 6, 2018, https://tech.newstatesman.com/business/russia-uk-critical-infrastructure. 42 “UK Political Parties Warned of Russian Hacking Threat: Report,” Reuters, March 12, 2017, https://www.reuters.com/article/us-britain-russia-cybercrime-idUSKBN16J0OE. 6
Cyber Governance and Digital Democracy 2019/1 manipulation attempts, than a direct hack. Thomas Rid’s disinformation attempts45. A number of such disinformation US Senate hearing demonstrated that both hacking and attempts involving the immigrants and immigrant-related disinformation operations from Russia and China during violence have become popular among German far-right the Brexit vote favored no particular candidate, but rather digital circles, perhaps the best-known case being the sought to exacerbate existing divisions and polarization ‘Lisa’ story46. Even after the exposure of the ‘Lisa story’ over the refugee problem, immigration and the political as a disinformation case, Russian-origin disinformation power balance between London and Brussels . Like the 43 campaigns on immigration and integration still remain US, Russian-origin disinformation in the UK has been easy popular among the German far-right47. To defend against to spot and map-out. University of Edinburgh researchers such attempts, political sides in Germany have entered have discovered that 3000 types of unique content can be into a trans-partisan agreement before the September traced to the Russian Internet Research Agency, with around 2017 election to refrain from exploiting each other’s’ leaked 150,000 unique accounts created by pro-Russian networks political data and significantly limit the use of Twitter bots to to post specifically on the Brexit referendum44. These boost the spread and engagement of their online political accounts were generally involved in anti-NATO and anti-EU messages48. Facebook contributed to the training of the content dissemination and focused on the British far-right political parties in securing digital infrastructure systems, audience through nationalist and isolationist content. as well as how to deal with digital disinformation as fast as possible, with the help of the voters and national fact- Germany on the other hand, has been targeted from checking initiatives49. German domestic intelligence agency a multitude of vulnerabilities, including its Bundestag BfV and Federal Office for Information Security, BSI, both took network, Ministry of Finance digital accounts, Ministry of an active part in training political parties against a number of Foreign Affairs records and the Christian Democratic Union vulnerabilities50. BSI even offered its cyber defense services (CDU) party infrastructure. Like France, Germany has to all parties51. Like France, Germany also set up a large been primarily targeted by the APT28 hacker team, a GRU network of citizen-led, autonomous fact-checking networks cyber-extension, as well as carefully curated and targeted to widen its disinformation defense capabilities. 43 Thomas Rid, “Disinformation: A Primer in Russian Active Measures and Influence Campaigns,” Pub. L. No. 33017, § Select Committee on Intelligence (2017). 44 Robert Booth et al., “Russia Used Hundreds of Fake Accounts to Tweet about Brexit, Data Shows,” The Guardian, November 14, 2017, sec. World news, https://www.theguardian.com/world/2017/nov/14/how-400-russia-run-fake-accounts-posted-bogus-brexit-tweets. 45 Carrel and Shalal, “Germany Says Its Government Computers Secure after ‘Isolated’ Hack.” 46 “The ‘Lisa Case’: Germany as a Target of Russian Disinformation,” NATO, 2016, http://www.nato.int/docu/review/2016/Also-in-2016/lisa-case-germany-target-russian-disinformation/EN/index.htm. 47 Constanze Stelzenmüller, “The Impact of Russian Interference on Germany’s 2017 Elections,” Brookings (blog), June 28, 2017, https://www.brookings.edu/testimonies/the-impact-of-russian-interference-on-germanys-2017-elections/. 48 Michael Schwirtz, “German Election Mystery: Why No Russian Meddling?,” The New York Times, January 20, 2018, sec. World, https://www.nytimes.com/2017/09/21/world/europe/german-election-russia.html. 49 Sara Germano, “Facebook, Germany to Collaborate Against Election Interference,” Wall Street Journal, January 20, 2019, sec. Business, https://www.wsj.com/articles/facebook-germany-to-collaborate-against-election-interference-11548004995. 50 Sumi Somaskanda, “The Cyber Threat To Germany’s Elections Is Very Real,” The Atlantic, September 20, 2017, https://www.theatlantic.com/international/archive/2017/09/germany-merkel-putin-elections-cyber-hacking/540162/. 51 “German Cyber Defense Blends Military and Commerce,” Deutsche Welle, 09 2018, https://www.dw.com/en/german-cyber-defense-blends-military-and-commerce/a-45636325. 7
Cyber Governance and Digital Democracy 2019/1 Charting the Pro-Russian Information Ecosystem in Turkey With such direct and easily identifiable Russian meddling in a norm, not an exception, which makes it harder to isolate the some of the most powerful nations of the West, researchers researched anomaly within a wider pool of other anomalies. have recently begun exploring how Russia-origin There is also the critical question of causality. Following the disinformation efforts shape the political debate in the rest of popularized cases in the US, UK, France and Germany, more the world. Turkey is a natural case study. After all, Turkey has countries have begun reporting cases of disinformation, even been a Cold War buffer country, lies right at the intersection when such cases have no measurable effect on any political of Western and Eastern security ecosystems and is adjacent outcome58. This availability bias is plaguing the field with an to three major civil wars – Iraq, Syria and Ukraine. It has long enormous volume of contextually irrelevant cases of fake been a strategically important country and lies adjacent to news that don’t spread beyond a very small network and/or some of the most problematic politically contested regions have no political or social implication59. More importantly, an of the Balkans, Caucasus, Middle East and North Africa. excessive focus on Russian disinformation generate myopia According to International Telecommunication Union (ITU), that overlooks the agency of domestic players in making their around 70% of the population has access to the Internet52 countries vulnerable to disinformation in general60. Today, it and according to the UK-based media analytics company is possible to locate Russian disinformation in a large number WeAreSocial, Turkey is one of the top countries in terms of countries, although Russia is by no means the only player of social media usage53. Despite restrictions, Turkey is still that weaponizes digital information or systematically hacks one of the most active countries in terms of discussion and international political actors. This creates an availability bias dissemination of political information online54, and ranks in the field, which distorts the extent to which Russia matters among the most active countries in terms of using social in digital space and substantially downplays the agency of media for political communication purposes . Yet, it is also 55 the pre-existing political and media actors and institutions, one of the most vulnerable countries to disinformation, bot analytically reducing them into non-entities. usage and cyber-attacks . Therefore, if there is an ideal 56 country to study the impact of disinformation on politics, This is a problematic way to approach disinformation, Turkey comes very close to that definition. because its conceptual root – propaganda – is certainly not new to political communication. States have been deploying Yet, Turkey is also a difficult country to study in terms of propaganda for centuries through the most recent and disinformation, because it is already plagued by high-levels widespread media outlets they could access in any given of fake news contamination57. The overall poor state of the era. Digital media is simply another step in the long history information environment in the country renders disinformation of propaganda and political diversion, spread in the past 52 “Measuring the Information Society Report 2018,” International Telecommunications Union, 2018, https://www.itu.int/en/ITU-D/Statistics/Pages/publications/misr2018.aspx. 53 “Digital in 2018: World’s Internet Users Pass the 4 Billion Mark,” We Are Social, January 30, 2018, https://wearesocial.com/blog/2018/01/global-digital-report-2018. 54 Erkan Saka, “Social Media in Turkey as a Space for Political Battles: AKTrolls and Other Politically Motivated Trolling,” Middle East Critique 27, no. 2 (April 3, 2018): 161–77, https://doi.org/10.1080/19436149.2018.1439271. 55 “Reuters Institute Digital News Report” (Oxford, UK: Reuters Institute, Oxford University, 2018), http://www.digitalnewsreport.org/. 56 Barçın Yinanç, “Poor Media Literacy ‘making Turks Vulnerable to Fake News’ - Turkey News,” Hürriyet Daily News, December 10, 2018, http://www.hurriyetdailynews.com/poor-media-literacy-making-turks-vulnerable-to-fake-news-139582. 57 Mark Lowen, “Hunting for Truth in a Land of Conspiracy,” BBC News, November 15, 2018, sec. Europe, https://www.bbc.com/news/world-europe-46137139. 58 Herman Wasserman, “Fake News from Africa: Panics, Politics and Paradigms,” Journalism, December 17, 2017, 1464884917746861, https://doi.org/10.1177/1464884917746861; Marco Visentin, Gabriele Pizzi, and Marco Pichierri, “Fake News, Real Problems for Brands: The Impact of Content Truthfulness and Source Credibility on Consumers’ Behavioral Intentions toward the Advertised Brands,” Journal of Interactive Marketing 45 (February 1, 2019): 99–112, https://doi.org/10.1016/j.intmar.2018.09.001; Edson C. Tandoc Jr, Zheng Wei Lim, and Richard Ling, “Defining ‘Fake News,’” Digital Journalism 6, no. 2 (February 7, 2018): 137–53, https://doi.org/10.1080/21670811.2017.1360143. 59 Dominic Spohr, “Fake News and Ideological Polarization: Filter Bubbles and Selective Exposure on Social Media,” Business Information Review 34, no. 3 (September 1, 2017): 150–60, https://doi.org/10.1177/0266382117722446; Tim Groeling, “Media Bias by the Numbers: Challenges and Opportunities in the Empirical Study of Partisan News,” Annual Review of Political Science 16, no. 1 (2013): 129–51, https://doi.org/10.1146/annurev-polisci-040811-115123. 60 Samantha Bradshaw and Philip N. Howard, “The Global Organization of Social Media Disinformation Campaigns,” Journal of International Affairs 71, no. 1.5 (2018): 23–32. 8
Cyber Governance and Digital Democracy 2019/1 through word of mouth, sealed envelopes, telegram, radio, language media outlet is Sputnik-Türkiye, which was one television and motion picture. Success of propaganda of the first foreign language branches of the agency after through all of these periods relied primarily on two factors: it replaced all previous Russian foreign-language services the perpetrator’s understanding of the media system, and in November 2014. Around the same time, Rusya’nın Sesi – the perpetrator’s successful diagnosis of the social divisions the primary pro-Russian news radio in Turkey – was named in a target country. Propaganda therefore, always depends RSFM and continued under the aegis of the Sputnik News on how well its wielder understands the social impact of Agency. Aydınlık is a well-known pro-Russian outlet, that the delivery mechanism, be it telegram, motion picture or regularly reports events directly related to Turkish-Russian Twitter, and how well it understands what makes its audience relations with an emphasis on the Russian view. There is tick. From this perspective, a propagandist’s success relies also a pro-Turkish-Russian relations website called TurkRus. largely on the pre-existing information environment and the com, run by journalist Suat Taşpınar, where news and extent of grievances within the audience. opinion on Turkish-Russian relations are shared frequently. Outside Sputnik-Türkiye, RSFM and Aydınlık however, In light of the growing criticism in the disinformation literature what constitutes a pro-Russian outlet in Turkey is a highly against studies that merely state that ‘disinformation exists’ contested and often a context-specific designation. Being or focus solely on Russian disinformation diffusion without considered ‘pro’ any foreign country, be it Russia, United any political and media environment context, this report States or otherwise, is generally considered libelous in aims to go one step further. Rather than only looking at Turkey. To that end, being called pro-Russian, like being whether there is Russian disinformation in Turkey (there called pro-American, is usually an external allegation to is), this study seeks to explore whether Russian information defame an actor and is also usually denied by the target(s). operations in Turkey matter, and have any influence on This complicates an analyst’s job, especially when the study Turkey’s wider information landscape. Do they influence is empirical, because the outlining the ‘pro-Russian media any mainstream conversation and have a measurable environment’ becomes a moving target. Regardless, and effect such as polarization and/or shifting election results, as demonstrated in this research report, all shades of the or do they merely exist in isolation, without any substantial Turkish media spectrum have published news reports and engagement and relevance? analyses that could be considered as ‘pro-Russian’ or ‘close to Putin’s position’, under different contexts and content. In To do that, this study dissects 3 of the most important events other words, the mainstream media in Turkey measurably in recent Turkish-Russian bilateral relations (downing of the shifts into a pro-Russian narrative on issues directly related SU24 jet, assassination of the Russian ambassador and the to Turkish-Russian relations. S400 negotiations) and arguably 2 of the most important domestic events in recent Turkish politics (2016 coup attempt Prominent studies61 that explore Russian disinformation in and 24 June 2018 general elections) and aims to measure the US or EU have so far identified ‘pro-Russian’ accounts in the impact of Russian-origin information campaigns against three ways: a) URL/domain root tracking to see if they lead to the wider Turkish information network in these most critical Sputnik, RT or other Russian-language media or website link, episodes. This method yields a more accurate perspective b) location information of the account(s) involved (most US to evaluate the severity of Russian information efforts in and UK disinformation studies traced pro-Russian networks Turkey compared to isolating low-engagement or marginally based on their stated physical location within Russian relevant content types that have no measurable effect on the territory), and c) dominant language of the text shared by general information network. the account (whether it is mostly in Russian). A combination of these methods usually yield a fairly reliable network of Mainstream pro-Russian information ecosystem in Turkey is Russian influence actors in an information ecosystem. The fairly straightforward. Russia’s primary mainstream Turkish first method can be applied to the Turkish case, as pro- 61 Savvas Zannettou et al., “Disinformation Warfare: Understanding State-Sponsored Trolls on Twitter and Their Influence on the Web,” arXiv:1801.09288 [Cs], January 28, 2018, http://arxiv.org/abs/1801.09288; Michael Jensen, “Russian Trolls and Fake News: Information or Identity Logics?,” Journal of International Affairs 71, no. 1.5 (2018): 115–24; Yevgeniy Golovchenko, Mareike Hartmann, and Rebecca Adler-Nissen, “State, Media and Civil Society in the Information Warfare over Ukraine: Citizen Curators of Digital Disinformation,” International Affairs 94, no. 5 (September 1, 2018): 975–94, https://doi.org/10.1093/ia/iiy148; Denis Stukal et al., “Detecting Bots on Russian Political Twitter,” Big Data 5, no. 4 (December 1, 2017): 310–24, https://doi.org/10.1089/big.2017.0038; Emilio Ferrara et al., “The Rise of Social Bots,” Communications ACM 59, no. 7 (June 2016): 96–104, https://doi.org/10.1145/2818717. 9
Cyber Governance and Digital Democracy 2019/1 Russian opinion actors in Turkey do share Sputnik or Russia Russia’s position. As the mainstream media environment Today news frequently. The other two however, are not very shifts gradually into a more pro-Russian tone, identifying pro- reliable in the Turkish context, because they can be easily Russian outlets and networks become even harder. The only faked. Like the well-known attribution problem in cyber- exceptions, of course are Sputnik-Türkiye, RSFM, because attacks (i.e. who really conducted the attack), location and they are explicitly Russian-owned and funded. The closest language-based attribution of information operations can domestic political outlet that comes closest to being an be masked, leading to false attribution. Furthermore, in our indigenous pro-Russian player is Aydınlık, given its frequent study, less than .0001% of the content (14 tweets only out explicit support for Russian policy. However, in light of the of 183 million) was sent by accounts that contains location new data presented in this study, it is more accurate to track information in Russia, or had Russian-language text in their ‘pro-Russian content’, rather than an outright ‘pro-Russian content data. outlet’, as the former can be traced across all shades of the Turkish media environment, whereas the latter becomes a Further complications arise from separating a pro-Russian subjective designation, open to debate. media outlet from an anti-Western media outlet. A range of Turkish outlets have been designated as ‘pro-Russian’ due From a methodological standpoint, all of this makes a truly to their anti-NATO and anti-EU reporting bias, although some objective computational, large-volume study on Russian of these outlets sometimes reject this designation, defining information operations in Turkey tricky. Instead of following themselves as ‘nationalist’. From an analyst’s perspective it the best-known mainstream studies that focus on Russian becomes further difficult to separate NATO, EU and Russia disinformation in the West through focusing on follower news reports of a nationalist or an Islamist outlet, given both networks and most commonly shared news domains, we forms of reporting are heavily against Western institutions had to go a step further. In this study, we use the follower and report Turkish-NATO and Turkish-EU cooperation in network and web URL tracing approaches, while adding overwhelmingly critical terms. This doesn’t necessarily make a third dimension: sentiment. We have trained our topic them pro-Russian, as most of those outlets are also often modelling algorithm in Turkish political text to identify critical of Moscow. Even further complicating the picture, positive and negative sentiment digital content related to and as demonstrated empirically in this research report, events and topics on Russian-Turkish relations. This three- in the last 3 years, most mainstream, high-circulation pro- layered strategy was necessary, as Russian information government and opposition news outlets alike have begun, domain is not nearly as explicit, straightforward and brazen at different times, reporting content that is aligned with as we observe in other Western cases. Methodology Data source. The data used in this study is generated via Facebook, which yielded a total of 1,163,856 words over 3 the Twitter streaming API. Our research group has built a years, Twitter gave us 486,052,996 words in total through the crawler that live-scrapes all tweets and their metadata since same time period, rendering Twitter a far better venue for this 24 November 2015 (SU-24 downing incident) that contain type of research. In our study on Turkish elections, we have the n-grams ‘Rus_’, ‘Putin_’ and ‘Moskova_’. These n-grams pre-designated 6 of the most-shared proven disinformation are derived from the keyword analysis tool KWFinder and cases that emerged during the election period. To explore topic-wise allow us to extract 99.998% of all Twitter content disinformation during the failed 2016 coup attempt, we related to Turkish-Russian relations in Turkish-language. similarly analyzed 5 of the most prominent cases of fake We don’t call our approach ‘stemming’, which is mostly news. These disinformation types have been exposed by understood as an acronym for n-grams. Stemming works Teyit.Org, a major Turkish fact-checking initiative. better with inflecting languages like English, whereas has a low reliability in agglutinative languages like Turkish. In Case selection. Our cases are selected due to their the latter, n-gram approach yielded a more reliable result significant digital popularity compared to other cases in compared to ‘stemming’. We primarily use Twitter in this Turkish-Russian relations, and their political impact on study, with periodic robustness checks on Facebook. In bilateral relations. 24 June elections were picked, as these comparison to our scraping effort on public accounts on were arguably the most important political election in 10
Cyber Governance and Digital Democracy 2019/1 Turkey, with the higher stakes and acute rivalry compared a regime-shift component. The importance of 24 June to previous elections. Most importantly, 24 June elections elections is evidenced by their popularity on social media. inaugurated the new political system in Turkey, and has Selected Cases Percentage of Dirty Data Clean Data S400 negotiations 3.937 44,394,129 (5 benchmarks – longitudinal) 2016 Coup attempt 21.593 27,459,214 Assassination of the Ambassador 11.353 18,667,492 SU24 downing incident 9.825 13,560,108 Discarded Cases Percentage of Dirty Data Clean Data Russia’s annexation of Crimea (Mar 2014) 15.024 1,901,403 TurkStream Negotiations + Signing (longitudinal) 2.079 871,031 White Helmets (longitudinal) 46.492 139,059 Akkuyu Nuclear Power Plant (longitudinal) 1.291 84,493 Selahattin Demirtaş visit to Moscow (Dec. 2015) 11.938 37,381 Table 1 - Selected and discarded cases based on the proportion of cleaned data and post-cleaning data size Selected Robustness Check Percentage of Dirty Data Clean Data 24 June 2018 Elections (Presidential + General) 17.884 79,823,491 Discarded Robustness Check Alternatives Percentage of Dirty Data Clean Data 2017 referendum 21.091 24,043,032 1 November 2015 Elections (General) 14.726 18,129,491 June 2015 Elections (General) 11.958 14,939,251 2014 Presidential elections 8.083 7,249,219 2014 local elections 6.179 3,140,402 Table 2 - Selected robustness check case and discarded alternatives based on the proportion of cleaned data and post-cleaning data size 11
Cyber Governance and Digital Democracy 2019/1 Data cleaning. In this study, we clear out a type of bot-driven follower-to-like ratio (an account that contains too many content commonly observed in Turkey: gibberish, randomly- posts with disproportionate likes and retweets compared to selected words that don’t contain any meaning. This type its follower count, there is a high likelihood that it is a bot). of automated content is easy to automatically detect and remove because The First Letters Of All Words In Such Automated Content/Sentiment Analysis. For automated text Tweets Are Capitalized. Often, bots that have nothing to analysis we use ReadMe: Software for Automated Content do with Turkish-Russian relations (such as sports-related Analysis66, CEM: Coarsened Exact Matching Software and or advertiser bots) are programmed to auto-select n-grams WhatIF: Software for Evaluating Counterfactuals, all of which based on their popularity, with no relation to any political we trained with Turkish-language official document, media incident. This results in quite a large volume of meaningless text and news website comments text specifically related data that skews the results of Turkey-based disinformation to Turkish-Russian relations and Turkish foreign policy over research, and thus, have to be cleaned. We don’t clean out the course of 6 months. Through random selection semi- tweets that are bot-driven, but have a meaningful sentence supervision tests, this heavy focus on texts related specifically structure. The total number of tweets we study in this report Turkish-Russian political relations has allowed us to reach is 183,904,434 post-cleaning. Percentages of dropped 98% reliability in detecting Turkish-language sentiment tweets during the cleaning phase in each case are listed in (including sarcasm) correctly67. In addition to specialized Tables 1 and 2. learning through focused text, we believe that this high level of reliability also owes to the linguistic properties of Turkish Bot designation. We use a bot-detection algorithm that auto- as an agglutinative language as opposed to English as an identifies suspected bot accounts by employing four of the inflecting language. most established methods in the methodological literature: a) friend-to-follower ratio (Wang et. al., 201062), number of Latent Dirichlet Allocation: LDA is a topic model type, tweets (Howard and Kollanyi, 201663), account creation which is a statistical text-mining method that auto-discovers date (Jones, 2017 ) and text duplicates (Thieltges et. al. 64 relevant topic clusters in a large body of text. Its relevance 201865). Only if an account posts content that meets one of algorithm is driven by Dirichlet distributions that are built as these criteria, we do a second robustness check through a) ‘topic-per-document’ and ‘words-per-topic’ classification. the use of URL shorteners (a main indicator of automation, Our pre-processing routine includes tokenization, removing because URL shorteners, like trib.al, bit.ly or tinyurl.com, words fewer than 3 words, removing stopwords, lemmatizing track traffic to a link), b) its tweet history of using more than words and reducing n-grams to their root form. In this study, 3 languages (bot accounts post automated messages in we use the LDA approach defined by McCallum (et. al. an average of 5-6 different languages to push a particular 200768). narrative in multiple linguistic and time-zone domains), c) 62 Alex Hai Wang, “Detecting Spam Bots in Online Social Networking Sites: A Machine Learning Approach,” in Data and Applications Security and Privacy XXIV, ed. Sara Foresti and Sushil Jajodia, Lecture Notes in Computer Science (Springer Berlin Heidelberg, 2010), 335–42. 63 Philip N. Howard and Bence Kollanyi, “Bots, #Strongerin, and #Brexit: Computational Propaganda During the UK-EU Referendum,” SSRN Scholarly Paper (Rochester, NY: Social Science Research Network, June 20, 2016), https://papers.ssrn.com/abstract=2798311. 64 Marc Owen Jones, “Hacking, Bots and Information Wars in the Qatar Spat,” Washington Post, June 7, 2017, https://www.washingtonpost.com/news/monkey-cage/wp/2017/06/07/hacking-bots-and-information-wars-in-the-qatar-spat/. 65 Andree Thieltges et al., “Effects of Social Bots in the Iran-Debate on Twitter,” arXiv:1805.10105 [Cs], May 25, 2018, http://arxiv.org/abs/1805.10105. 66 Daniel J. Hopkins and Gary King, “A Method of Automated Nonparametric Content Analysis for Social Science,” American Journal of Political Science 54, no. 1 (2010): 229–47, https://doi.org/10.1111/j.1540-5907.2009.00428.x. 67 Peng Liu et al., “Sarcasm Detection in Social Media Based on Imbalanced Classification,” in Web-Age Information Management, ed. Feifei Li et al., Lecture Notes in Computer Science (Springer International Publishing, 2014), 459–71. 68 A. McCallum, X. Wei, and X. Wang, “Topical N-Grams: Phrase and Topic Discovery, with an Application to Information Retrieval,” in Seventh IEEE International Conference on Data Mining (ICDM 2007)(ICDM), 2007, 697–702, https://doi.org/10.1109/ICDM.2007.86. 12
You can also read